PROJECT TITLE :

JiTTree: A Just-in-Time Compiled Sparse GPU Volume Data Structure

ABSTRACT:

Sparse volume data structures enable the efficient representation of large however sparse volumes in GPU memory for computation and visualization. But, the selection of a particular data structure for a given knowledge set depends on several factors, like the memory budget, the sparsity of the data, and knowledge access patterns. Generally, there's no single optimal sparse knowledge structure, however a set of several candidates with individual strengths and downsides. One answer to the present drawback are hybrid knowledge structures which locally adapt themselves to the sparsity. But, they sometimes suffer from increased traversal overhead which limits their utility in many applications. This paper presents JiTTree, a novel sparse hybrid volume knowledge structure that uses simply-in-time compilation to beat these issues. By combining multiple sparse knowledge structures and reducing traversal overhead we leverage their individual advantages. We demonstrate that hybrid knowledge structures adapt well to a large range of information sets. They're particularly superior to different sparse information structures for data sets that locally vary in sparsity. Attainable optimization criteria are memory, performance and a combination thereof. Through simply-in-time (JIT) compilation, JiTTree reduces the traversal overhead of the resulting optimal information structure. So, our hybrid volume knowledge structure permits efficient computations on the GPU, whereas being superior in terms of memory usage compared to non-hybrid information structures.